Make Pets Found (using AI) - Day 4

3 minute read

Day 4

Into our fourth full day and we were really on a roll. The slightly inclement weather forced us to stay inside for most of the day, spread across the kitchen and living room in the familiar position of hunched-over-laptop.

Jesús worked on improving and deploying his colour classification app which will help feed into a model he’s building to identify cat breed colours. To help this, Matt and Laura (AKA the fur colour classification team) spent time grouping individual colours into ‘cat colour’ categories. For example, a cat colour category might be ginger, so they grouped any hex codes that might be considered ginger - ie oranges, yellows, creams etc.

Sam focused on using the object detection provided by AWS SageMaker to train a model to recognise breeds of dogs and provide dimensions for a bounding box to our API to feed into other models for colour/pattern detection, and help to create lost pet posters.

On the backend, George and Luke worked on using Fuse.js to build a weighted search system which will let us scan our app for similar images (which our machine learning model provides tags for). This allows users to search our app for their beloved pet, and we’ll return similar matches if Tiddles has already been found!

James continued working on the frontend to ensure that there’s an interface to use the backend API.

After another culinary delight courtesy of Laura (lamb something or other), we had a team talk about our proposed trek/trip to Granada the next day. Tradition on a Gravitywell hackathon, we’d meet in the kitchen at 8am to prepare sandwiches for our ‘death march’. Some of the team (mainly George) enjoyed an ABV upgrade to their after dinner drinks in the form of rum and coke, but the night was an impressively early one. After all, we’d be scaling a mountain in a matter of hours…